[Asterisk-Users] Re: Generic X100P's

[Peter Svensson]

On Tue, 12 Oct 2004, Stephen R. Besch wrote:

> > That is pretty much what I said. The lack of buffering is a feature - 
> > latency is the enemy in telecomunications. That is why I said 100-200us 
> > should be acceptable. This should be redily aceivable, even with shared 
> > interrupts.
> > 
> 200us probably would be. And, you could probably get average interrupt 
> latency that low. The problem is that in any real machine, even when 
> operating in real time mode, maximum interrupt latency is much longer - 
> sometimes in the ms range - and the standard deviation isn't that great 
> either. If you don't believe it, hook a scope up to one of the active 
> interrupt request lines on the bus. It's very revealing.

There really cannot be that much of a problem - it works! We see a few 
missed interrupts a day on our TE405P and the machine is really stock. 
>From my experience the jitter on the interrupts is really low. 

We used to run a user-space based app that bit-banged a can bus at 1kHz.  
No misses were allowed or the log would have to be restarted. This worked
even with X running. From user-space. There were no real problems, and
after a few tewak basicall no missed deadlines. Of course, any broken
driver that took a long time to service its interrupts would have left us 
dead in the waters. I'm just saying we never had a problem. The jitter was 
on the order of 100us or so.

Anyway, shared or not shared interrupts should not have a great effect on 
servicing interrupts. No unless either of the two drivers are taking a 
long time just to determine that they were not the intended receiver of 
the interrupt.

For even stricter timings there is the realtime interrupts from 
http://linuxdevices.com/articles/AT6105045931.html. 5 us worst case. Nice. 
:-)

Peter